Introduction to PCL:
The Point Cloud Library
1 - basic topics
Alberto Pretto
Thanks to Radu Bogdan Rusu, Bastian Steder
and Jeff Delmerico for some of the slides!
Contact
Alberto Pretto, PhD
Assistant Professor
Location: DIAG A209
[email protected]
Point clouds: a definition
A point cloud is a data structure used to represent a
collection of multi-dimensional points and is
commonly used to represent three-dimensional data.
The points usually represent the X, Y, and Z
geometric coordinates of a sampled surface.
Each point can hold additional information: RGB
colors, intensity values, etc...
Where do they come from?
2/3D Laser scans
Laser triangulation
Stereo cameras
RGB-D cameras
Time of flight cameras
Simulation
...
Point clouds in robotics
Navigation / Obstacle avoidance
Object recognition and registration
Grasping and manipulation
Point Cloud Library
→ pointclouds.org
The Point Cloud Library (PCL) is a standalone, large scale, open source
(C++) library for 2D/3D image and point cloud processing.
PCL is released under the terms of the BSD license, and thus free for
commercial and research use.
PCL provides the 3D processing pipeline for ROS, so you can
also get the perception pcl stack and still use PCL standalone.
Among others, PCL depends on Boost, Eigen, OpenMP,...
PCL Basic Structures: PointCloud
A PointCloud is a templated C++ class which basically
contains the following data fields:
– width (int) - specifies the width of the point cloud dataset in
the number of points.
→ the total number of points in the cloud (equal with the
number of elements in points) for unorganized datasets
→ the width (total number of points in a row) of an organized
point cloud dataset
– height (int) - Specifies the height of the point cloud dataset
in the number of points
→ set to 1 for unorganized point clouds
→ the height (total number of rows) of an organized point
cloud dataset
– points (std::vector <PointT> ) - Contains the data array
where all the points of type PointT are stored.
–
PointCloud vs. PointCloud2
We distinguish between two data formats
for the point clouds:
– PointCloud<PointType> with a specific data type
(for actual usage in the code)
– PointCloud2 as a general representation containing
a header defining the point cloud structure (e.g., for
loading, saving or sending as a ROS message)
Conversion between the two frameworks is easy:
– pcl::fromROSMsg and pcl::toROSMsg
Important: clouds are often handled using smart
pointers, e.g.:
– PointCloud<PointType> :: Ptr cloud_ptr;
Point Types
PointXYZ - float x, y, z
PointXYZI - float x, y, z, intensity
PointXYZRGB - float x, y, z, rgb
PointXYZRGBA - float x, y, z, uint32 t rgba
Normal - float normal[3], curvature
PointNormal - float x, y, z, normal[3], curvature
→ See pcl/include/pcl/point_types.h for more
examples.
Building PCL Stand-alone Projects
# CMakeLists.txt
project( pcl_test )
cmake_minimum_required (VERSION 2.8)
cmake_policy(SET CMP0015 NEW)
find_package(PCL 1.7 REQUIRED )
add_definitions(${PCL_DEFINITIONS})
include_directories(... ${PCL_INCLUDE_DIRS})
link_directories(... ${PCL_LIBRARY_DIRS})
add_executable(pcl_test pcl_test.cpp ...)
target_link_libraries( pcl_test${PCL_LIBRARIES})
PCL structure
PCL is a collection of smaller, modular C++ libraries:
– libpcl_features: many 3D features (e.g., normals and curvatures, boundary
points, moment invariants, principal curvatures, Point Feature Histograms
(PFH), Fast PFH, ...)
– libpcl_surface: surface reconstruction techniques (e.g., meshing, convex hulls,
Moving Least Squares, ...)
– libpcl_filters: point cloud data filters (e.g., downsampling, outlier removal,
indices extraction, projections, ...)
– libpcl_io: I/O operations (e.g., writing to/reading from PCD (Point Cloud Data)
and BAG files)
– libpcl_segmentation: segmentation operations (e.g.,cluster extraction,
Sample Consensus model fitting, polygonal prism extraction, ...)
– libpcl_registration: point cloud registration methods (e.g., Iterative Closest
Point (ICP), non linear optimizations, ...)
– libpcl_range_image: range image class with specialized methods
It provides unit tests, examples, tutorials, ...
Point Cloud file format
Point clouds can be stored to disk as files, into the PCD
(Point Cloud Data) format:
– # Point Cloud Data ( PCD ) file format v .5
FIELDS x y z rgba
SIZE 4 4 4 4
TYPE F F F U
WIDTH 307200
HEIGHT 1
POINTS 307200
DATA binary
...<data>...
Funtions: pcl::io::loadPCDFile and pcl::io::savePCDFile
Example: create and save a PC
#include<pcl/io/pcd_io.h>
#include<pcl/point_types.h>
//....
pcl::PointCloud<pcl::PointXYZ>:: Ptr cloud_ptr (new pcl::PointCloud<pcl::PointXYZ>);
cloud->width=50;
cloud->height=1;
cloud->isdense=false;
cloud->points.resize(cloud.width∗cloud.height);
for(size_t i=0; i<cloud.points.size(); i++){
cloud->points[i].x=1024∗rand()/(RANDMAX+1.0f);
cloud->points[i].y=1024∗rand()/(RANDMAX+1.0f);
cloud->points[i].z=1024∗rand()/(RANDMAX+1.0f);
}
pcl::io::savePCDFileASCII("testpcd.pcd",*cloud);
Visualize a cloud
viewer->setBackgroundColor (0, 0, 0);
viewer->addPointCloud<pcl::PointXYZ> ( in_cloud, cloud_color,
"Input cloud" );
viewer->setPointCloudRenderingProperties
(pcl::visualization::PCL_VISUALIZER_POINT_SIZE, 1, "Input cloud");
viewer->initCameraParameters ();
viewer->addCoordinateSystem (1.0);
while (!viewer->wasStopped ()) viewer->spinOnce ( 1 );
Basic Module Interface
Filters, Features, Segmentation all use the
same basic usage interface:
– use setInputCloud() to give the input
– set some parameters
– call compute() or filter() or align() or ... to get the
output
PassThrough Filter
Filter out points outside a specified range in one
dimension.
–
pcl::PassThrough<T> pass_through;
pass_through.setInputCloud (in_cloud);
pass_through.setFilterLimits (0.0, 0.5);
pass_through.setFilterFieldName ("z");
pass_through.filter( *cutted_cloud );
Downsampling
Voxelize the cloud to a 3D grid. Each occupied
voxel is approximated by the centroid of the
points inside it.
–
pcl::VoxelGrid<T> voxel_grid;
voxel_grid.setInputCloud (input_cloud);
voxel_grid.setLeafSize (0.01, 0.01, 0.01);
voxel_grid.filter ( *subsamp_cloud ) ;
Features example: normals
pcl::NormalEstimation<T, pcl::Normal> ne;
ne.setInputCloud (in_cloud);
pcl::search::KdTree<pcl::PointXYZ>::Ptr tree (new
pcl::search::KdTree<pcl::PointXYZ> ());
ne.setSearchMethod (tree);
ne.setRadiusSearch (0.03);
ne.compute ( *cloud_normals );
Segmentation example
A clustering method divides an unorganized point
cloud int into smaller, correlated, parts.
EuclideanClusterExtraction uses a distance
threshold to the nearest neighbors of each point to
decide if the two points belong to the same cluster.
– pcl::EuclideanClusterExtraction<T> ec;
ec.setInputCloud (in_cloud);
ec.setMinClusterSize (100);
ec.setClusterTolerance (0.05); // distance threshold
ec.extract (cluster_indices);
Iterative Closest Point - 1
ICP iteratively revises the transformation
(translation,rotation) needed to minimize the
distance between the points of two raw scans.
– Inputs: points from two raw scans, initial
estimation of the transformation, criteria for
stopping the iteration.
– Output: refined transformation.
Iterative Closest Point - 2
The algorithm steps are :
– 1. Associate points of the two cloud using the
nearest neighbor criteria.
– 2. Estimate transformation parameters using a
mean square cost function.
– 3. Transform the points using the estimated
parameters.
– 4. Iterate (re-associate the points and so on).
Iterative Closest Point - 3
IterativeClosestPoint<PointXYZ, PointXYZ> icp;
// Set the input source and target
icp.setInputCloud (cloud_source);
icp.setInputTarget (cloud_target);
// Set the max correspondence distance to 5cm
icp.setMaxCorrespondenceDistance (0.05);
// Set the maximum number of iterations (criterion 1)
icp.setMaximumIterations (50);
// Set the transformation epsilon (criterion 2)
icp.setTransformationEpsilon (1e-8);
// Set the euclidean distance difference epsilon (criterion 3)
icp.setEuclideanFitnessEpsilon (1);
// Perform the alignment
icp.align (cloud_source_registered);
// Obtain the transformation that aligned cloud_source to cloud_source_registered
Eigen::Matrix4f transformation = icp.getFinalTransformation ();